fix(#380): tolerate Claude harness output-scan markers in strict JSON gates

[azalio]

Summary

Fixes #380.

Claude Code v2.1.210+ may prepend a marker line like [harness: subagent output matched instruction-shaped pattern(s): ...] to a subagent report when the output matches instruction-shaped patterns. The payload itself is the original valid agent JSON. detect_truncated_agent_output previously treated any leading text as a "wrapped in prose" signal and returned truncated: true, causing valid Monitor / Predictor / Evaluator / Actor payloads to be spuriously rejected — wasting a retry or stopping the workflow with a misleading JSON-format error.

Changes

• _HARNESS_MARKER_PREFIX — module-level constant documenting the exact documented marker form
• _strip_harness_markers(text) — helper that removes only lines whose stripped form starts with that prefix; arbitrary prose is left intact so the downstream parser still rejects it
• detect_truncated_agent_output — calls the helper before JSON parsing (after the empty-response check); adds harness_markers_stripped: list[str] to every return dict so the event is observable
• 6 new regression tests covering: monitor + actor payloads with a harness marker, multiple markers, arbitrary prose still rejected, marker followed by trailing prose still rejected, clean JSON produces empty list
• make render-templates propagated the change to all generated trees (.map/scripts/, src/mapify_cli/templates/, .claude/ dev copy)

Test plan

• pytest tests/test_map_step_runner.py::TestDetectTruncatedAgentOutput — 28/28 pass (22 existing + 6 new)
• make check — 3802 passed, 4 skipped, ruff/mypy/pyright all clean
• make check-render — generated trees match source

---

Generated by Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Improved parsing of agent responses that begin with recognized harness marker lines.
  • Prevented valid JSON responses from being incorrectly flagged as truncated when these markers are present.
  • Continued to flag unexpected leading prose as malformed or truncated.

• Tests

  • Added coverage for single and repeated harness markers, valid responses, and invalid leading text.
  • Responses now report which harness markers were removed during processing.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 842fe378-0258-4aa5-9380-17227c9f2966

📥 Commits

Reviewing files that changed from the base of the PR and between cd00d02 and aa9fdf2.

📒 Files selected for processing (4)

• .map/scripts/map_step_runner.py
• src/mapify_cli/templates/map/scripts/map_step_runner.py
• src/mapify_cli/templates_src/map/scripts/map_step_runner.py.jinja
• tests/test_map_step_runner.py

---

📝 Walkthrough

Walkthrough

Changes

Harness marker tolerance

Layer / File(s)	Summary
Marker normalization and result propagation .map/scripts/map_step_runner.py, src/mapify_cli/templates/..., src/mapify_cli/templates_src/...	Leading Claude harness marker lines are stripped before JSON parsing, recorded in harness_markers_stripped, and propagated through detector return paths.
Marker parsing regression coverage tests/test_map_step_runner.py	Tests cover valid JSON with one or more markers, clean JSON, and rejection of arbitrary leading prose.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant AgentOutput
  participant detect_truncated_agent_output
  participant _strip_harness_markers
  participant JSONParser
  AgentOutput->>detect_truncated_agent_output: provide agent output
  detect_truncated_agent_output->>_strip_harness_markers: strip leading harness markers
  _strip_harness_markers-->>detect_truncated_agent_output: cleaned text and stripped markers
  detect_truncated_agent_output->>JSONParser: parse cleaned text
  JSONParser-->>detect_truncated_agent_output: parsed payload or validation result
  detect_truncated_agent_output-->>AgentOutput: return truncation result and stripped markers

Loading

Poem

I’m a rabbit with markers tucked neatly away,
JSON hops through the gate without delay.
Prose still gets stopped at the strict little door,
Clean results report what was stripped before.
Thump-thump—tests keep the contract in store!

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/compassionate-cerf-ooawte

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}